Ecological Indicators 140 (2022) 109041

Contents lists available at ScienceDirect

Ecological Indicators
journal homepage: www.elsevier.com/locate/ecolind

Original Articles

Mapping algal bloom dynamics in small reservoirs using Sentinel-2 imagery

in Google Earth Engine
Chippie Kislik a, Iryna Dronova a, Theodore E. Grantham a, Maggi Kelly a, b, *
a
Department of Environmental Sciences, Policy and Management, University of California, Berkeley, CA 94720-3114, USA
b
Division of Agriculture and Natural Resources, University of California, Berkeley, CA 94720-3114, USA

A R T I C L E I N F O A B S T R A C T

Keywords: Freshwater algal blooms have caused ecological damage and public health concerns throughout the world.
Chlorophyll-a Monitoring such blooms via in situ sampling is both costly and time-consuming, and satellite imagery provides a
Cyanobacteria rapid and relatively inexpensive way to supplement these techniques. Sentinel-2 MultiSpectral Imager data have
Time series
effectively detected chlorophyll-a, a proxy for algal biomass, in large bodies of water, but few studies have shown
Inland waters
the applicability in small (<10 km2) reservoirs, which are critically important for aquatic species, drinking water,
Klamath
Dam removal irrigation, cultural activities, and recreation. This study provides a test of the use of Sentinel-2 imagery in Google
Earth Engine for algal bloom detection in two small freshwater reservoirs in northern California, USA, from
October 2015 to December 2020. Google Earth Engine’s cloud computing allows for the analysis of extensive
datasets and time series, expanding the capacity to analyze the spatial and temporal heterogeneity of floating
algal blooms. Here we analyzed four spectral indices - Normalized Difference Vegetation Index (NDVI),
Normalized Difference Chlorophyll Index (NDCI), B8AB4, and B3B2 - to retrieve chlorophyll-a data for algal
bloom identification in two highly dynamic freshwater systems. We assessed the relationship between spectral
indices and monthly in situ water samples that were collected at three sites within the reservoirs using cubic
polynomial regression equations. NDCI, which leverages the red-edge wavelength, most accurately identified
chlorophyll-a across all study sites (highest adjusted R2 = 0.84, lowest RMSE = 0.02 µg/l), followed by NDVI. We
demonstrate that Sentinel-2 imagery can capture greater spatial and temporal heterogeneity of algal blooms than
typical in situ sampling. This suggests that remote sensing may be an increasingly important tool in monitoring
algal bloom dynamics in small reservoirs and other aquatic environments.

1. Introduction Furthermore, consuming even small amounts of toxins can be life-

threatening (Chorus et al., 2000). Blooms are increasing in prevalence
Freshwater algal blooms, defined as proliferations of toxic or non- in small lakes and reservoirs around the world (Brooks et al., 2016;
toxic algae that negatively affect freshwater ecosystems (Lopez et al., Kenneth Hudnell, 2008). In fact, lakes and reservoirs often experience
2008), impact every continent on the planet (Clark et al., 2017). the most severe and frequent algal blooms of all aquatic environments
Although algae occur naturally and provide nutritional and structural because of high nutrient influx from upstream rivers (Paerl et al., 2018).
benefits to aquatic ecosystems, proliferations can become a nuisance or While much algal bloom research has been conducted in marine eco­
even dangerous to animal and human health (Paerl et al., 2001). Such systems and large freshwater ecosystems, few researchers have studied
health effects include nausea, respiratory illnesses, liver problems, skin these events in small lakes and reservoirs despite the importance of these
damage (Davidson et al., 2014), and mortality of livestock, dogs, and water bodies for irrigation, energy usage, and water supply (Lobo et al.,
other animals that drink contaminated water (Backer et al., 2013; Harke 2021). Evaluating the spatio-temporal trends and variable dynamics of
et al., 2016). In addition to toxigenic events, algal blooms can cause a algal blooms in small lakes and reservoirs can provide public health
variety of ecological and economic problems, such as creating anoxic officials and aquatic managers with targeted information to mitigate
zones, clogging fish gills, producing strong odors, and impeding potentially harmful events around the globe (Anderson et al., 2012).
waterfront recreation (Hallegraeff, 1993; Weirich and Miller, 2014). Algal blooms are typically monitored through in situ methods such as

* Corresponding author.
E-mail address: [email protected] (M. Kelly).

https://doi.org/10.1016/j.ecolind.2022.109041
Received 21 March 2022; Received in revised form 23 May 2022; Accepted 4 June 2022
Available online 15 June 2022
1470-160X/© 2022 The Authors. Published by Elsevier Ltd. This is an open access article under the CC BY license (http://creativecommons.org/licenses/by/4.0/).
C. Kislik et al. Ecological Indicators 140 (2022) 109041

Fig. 1. Study sites of Iron Gate Reservoir and Copco Reservoir within the Klamath River Basin in northern California. There are a total of three in situ sample
locations (Copco, Below Copco, Iron Gate), each encircled by a 15-meter buffer representing locations of Sentinel-2 satellite data acquisition.

laboratory analysis of water samples that quantify concentrations of 2012). The red-edge and NIR wavelengths are advantageous for algal
algal toxins and chlorophyll-a (chl-a) (Table A.1), a proxy for algal bloom detection because, as with land vegetation reflectance patterns,
biomass (Caballero et al., 2020). However, in situ monitoring techniques floating algae have peak reflectance in the NIR and red-edge (Hu, 2009)
are time-consuming and require specialized laboratory equipment and high absorption in the red wavelengths, especially compared to
(Karlson et al., 2010), reducing the ability to collect data more surrounding water (Alawadi, 2010). However, the platforms that have
frequently than once a month per location. Since bloom cycles are het­ tested red-edge and NIR indices are primarily moderate and low reso­
erogeneous and can last from hours to days in various quantities (Cloern, lution satellite missions that are too coarse to map small lakes and res­
1996), it is important that monitoring techniques capture the rapidly- ervoirs. Higher resolution satellite images are needed for this pursuit.
changing dynamics of bloom events. Remote sensing data from satel­ Missions such as Landsat (30 m spatial resolution) and Sentinel-2
lites, aircraft, and unoccupied aerial vehicles (Kislik et al., 2018; Stumpf (10–60 m spatial resolution) offer finer-scale spatial data and may
and Tomlinson, 2007) have been used to complement in situ monitoring allow for more precise quantification of algal biomass in small fresh­
and improve our ability to observe blooms at broad spatial and temporal water lakes and reservoirs (Bresciani et al., 2018). Sentinel-2 Multi­
scales. Additionally, the growing availability of long-term, high-fre­ Spectral Imager (MSI) has emerged as a new leader in the field of algal
quency satellite imagery has enabled robust comparisons of spectral bloom remote sensing because of its spectral variety in red-edge and NIR
images with in situ data to develop models to forecast blooms (Mishra wavelengths, high temporal resolution, and high spatial resolution
et al., 2021; Ranjbar et al., 2021; Saberioon et al., 2020; Weber et al., (Ansper and Alikas, 2018; Dogliotti et al., 2018). Although Sentinel-2
2020). Chl-a is most commonly used to detect the presence of a variety sensors were originally designed for land-use applications (Cillero Cas­
of algal blooms using both in situ and remote sensing techniques because tro et al., 2020), this imagery has been applied to several algal bloom
it is easy to measure visually, although it is only a proxy and cannot studies involving coastal regions (Pahlevan et al., 2017) and large inland
determine algae to the species level (Kudela et al., 2017; Stumpf and lakes (Bramich et al., 2021; Ha et al., 2017). This satellite has been used
Tomlinson, 2007). Therefore, remote sensing techniques combined with in previous algal bloom research, yet few investigations have shown the
in situ water sampling measurements provides a robust mechanism to applicability of Sentinel-2 imagery in small reservoirs. The temporal
characterize the spatial and temporal dynamics of algal blooms. frequency of Sentinel-2 observations, which is higher than Landsat’s 16-
Satellite remote sensing has been used for decades to identify algal day revisit time (Caballero et al., 2020), provides an additional advan­
blooms, and advances in sensors have increased the ability to monitor tage for monitoring algal blooms in small bodies of water at repeat in­
blooms at finer spatial scales. Satellite imagery is particularly useful for tervals. It is important to track when and where blooms in small (<10
bloom detection because algal blooms typically float at the surface and km2) reservoirs occur to monitor public health concerns in these envi­
have distinct colorations, making them identifiable from space (Mat­ ronments because even small concentrations of toxic algae in small
thews et al., 2012). Previous studies have detected blooms using mul­ bodies of water can injure animals and humans (Hallegraeff et al., 2004;
tispectral imagery from SeaWiFS (1100 m spatial resolution), MERIS Laughrey et al., 2021). Furthermore, small reservoirs are critical for
(300 m spatial resolution), MODIS (250 m spatial resolution), and irrigation, water supply, energy production, recreation, and cultural
Sentinel-3 (300 m spatial resolution), leveraging the red-edge and near- usage (Ho and Goethals, 2019). Together, in situ measurements paired
infrared (NIR) wavelengths to identify chl-a and general greenness of with remote sensing data offer a unique window into the time and
blooms (Clark et al., 2017; Sebastiá-Frasquet et al., 2020; Shen et al., location of algal blooms in small bodies of water, helping public health

2
C. Kislik et al. Ecological Indicators 140 (2022) 109041

Fig. 2. Workflow diagram describing the data acquisition (in situ and satellite imagery), satellite image processing and analysis, and regression and error assessment
components (in situ and satellite imagery).

agencies and water quality managers address these events in a timely Aphanizomenon flos-aquae, particularly in the summer and fall months
and targeted manner. (Otten et al., 2015). Microcystis blooms in these two reservoirs have been
To test whether Sentinel-2 imagery can monitor algal bloom occur­ documented each year since 2004 (Asarian and Kann, 2011), producing
rences in small water bodies, we compared four spectral indices with in microcystin toxin levels that often exceed public health standards (Otten
situ water quality data from two small (10 km2 total) reservoirs in the et al., 2015). These toxic, recurring events threaten humans and en­
Klamath River Basin in northern California, USA. Specifically, we used dangered fish species, including chinook salmon (Oncorhynchus tsha­
chl-a and cyanobacteria toxin in situ data to evaluate the accuracy of wytscha), coho salmon (Oncorhynchus kisutch), and steelhead trout
satellite imagery for detecting blooms and to establish baseline water (Oncorhynchus mykiss) in the Klamath River (Kann and Corum, 2006; U.
quality data before several large dams in the system are removed in S. Department of the Interior and U.S. Department of Commerce, Na­
2023 (Klamath River Renewal Corporation, n.d.). This is the first study tional Marine Fisheries Service, 2013).
to utilize Sentinel-2 imagery in Google Earth Engine to analyze a long-
term (5-year) time series of algal bloom dynamics in small reservoirs.
2.2. In situ data collection
Our overall goal is to provide a refined methodology to monitor highly
localized and potentially toxic blooms that can be applied to other
PacifiCorp (an electric power company that operates hydroelectric
important aquatic habitats worldwide. Specific objectives for this study
dams in the Pacific Northwest) collects and publishes water quality data
include: (1) identifying which satellite-based spectral index best detects
pertaining to the Klamath River main stem and reservoirs, including
chl-a in two reservoirs of the Klamath River from 2015 to 2020; (2)
Copco and Iron Gate. We obtained PacifiCorp in situ water quality
understanding the temporal trends of algal blooms in these bodies of
sampling data at three locations: the outlet of Copco Reservoir (“Copco”)
water; and (3) quantifying spatial bloom dynamics and anomalies
(41.98◦ , − 122.331208998◦ ), the inlet of Iron Gate Dam (“Below
(higher and lower than average values) during this study period. This
Copco”) (41.97◦ , − 122.36438908◦ ), and the outlet of Iron Gate Dam
paper assesses the utility of high-resolution satellite imagery for small
(“Iron Gate”) (41.94◦ , − 122.432227627◦ ) (Fig. 1). These data included
algal bloom detection to supplement sampling techniques and enhance
chl-a concentrations and levels of microcystin, an acute liver toxin
bloom identification and monitoring.
(Rastogi et al., 2014), that were measured in water samples collected at
0.5 m depth and subsequently analyzed in a laboratory using Environ­
2. Materials & methods
mental Protection Agency standards (Watercourse Engineering, Inc.,
2020). The samples were collected at approximately monthly intervals
2.1. Study sites
between March and November from 2015 to 2020 (Fig. A.1). In situ
chlorophyll-a and microcystin toxin concentrations are highly corre­
The study sites for this research are the two most downstream res­
lated (adj. R2 > 0.85) at each of the 3 sites in this study (Fig. A.2).
ervoirs of the Klamath River in northern California near the Oregon
border: Copco Reservoir (41.97941◦ , − 122.304◦ ) and Iron Gate reser­
voir (41.93389◦ , − 122.435◦ ) (Fig. 1). These reservoirs are small 2.3. Sentinel-2 data collection & processing
(roughly 4 km2 surface area each) and hold about 34,000–59,000 acre-
feet of water (Klamath River Renewal Corporation, n.d.). Three dams We obtained multispectral imagery over the study period at the
were built within Copco and Iron Gate reservoirs: Copco I in 1918 (the sampling locations from Sentinel-2. These provide high-resolution (10 to
first blockade to migratory fish species in the Klamath River), Copco II in 20 m) images of the sampling sites at 5-day intervals when combining
1925, and Iron Gate in 1962. The dams were constructed for agricultural data from both Sentinel-2A, which began operating in 2015, and
water acquisition, hydroelectric production, and flow regulation (U.S. Sentinel-2B, in operation since 2017 (Caballero and Stumpf, 2020).
Department of the Interior and U.S. Department of Commerce, National After we acquired in situ water quality data, we processed Sentinel-2
Marine Fisheries Service, 2013). Both reservoirs are characterized by Level 1C imagery in Google Earth Engine (GEE), a satellite imagery
elevated nitrogen and phosphorus concentrations, warm water tem­ analysis application programming interface that enables global-scale
peratures, and high water residence times (Bozarth et al., 2010; Oliver processing (Fig. 2). Recent studies have proven the applicability of
et al., 2014). Their high nutrient levels are primarily attributed to in­ Google Earth Engine for water quality monitoring (Jia et al., 2019; Lobo
flows from the Upper Klamath Lake, a hypertrophic lake surrounded by et al., 2021; Maeda et al., 2019; Vaičiūtė et al., 2021; Wang et al., 2020;
intensive agriculture (Walker et al., 2012), but also to nitrogen fixation, Weber et al., 2020; Zong et al., 2019). We selected images from October
atmospheric deposition, and nutrient influxes from bottom sediments 2015 to December 2020 that were captured within 5 days of monthly in
(Asarian et al., 2009; Moisander et al., 2009). These conditions support situ water quality collection dates. About two-thirds of these images
the proliferation of algae, dominated by Microcystis aeruginosa and were within 0 to 2 days of an in situ collection date (Copco: 20 out of 30
total, Below Copco: 22 out of 32 total, Iron Gate: 20 out of 30 total). We

3
C. Kislik et al. Ecological Indicators 140 (2022) 109041

Table 1
Sentinel-2 spectral indices used for chlorophyll-a identification. The indices used in this study include the Normalized Difference Vegetation Index (NDVI), the
Normalized Difference Chlorophyll Index (NDCI), Band 8A Band 4 (B8AB4), and Band 3 Band 2 (B3B2). Rs indicates reflectance values from the satellite. The
wavelengths represented include the Near-Infrared (NIR), Red, Vegetation Red Edge 1 (Veg Red Edge 1), Narrow Near-Infrared (Narrow NIR), Green, and Blue.
Algorithm Equation Wavelengths Sentinel-2 Bands Reference

NDVI Rrs842 − Rrs665 NIR,Red (B8-B4)/(B8 + B4)

Ma et al., 2021
Rrs842 + Rrs665
NDCI Rrs705 − Rrs665 Veg red Edge 1, Red (b5-b4)/(B5 + B4)
Caballero et al 2020
Rrs705 + Rrs665
B8AB4 Rrs865 − Rrs665 Narro NIR, Red (B8A-B4)/(B8A + B4)
Khalili & Hasanlou 2019
Rrs865 + Rrs665
B3B2 Rrs560 − Rrs490 Green, Blue (B3-B2)/(B3 + B2) Cillero-Castro et al., 2020
Rrs560 + Rrs490

then implemented a cloud mask to all images using the QA60 (quality convert imagery from top of atmosphere to surface reflectance values.
assurance) band to remove thick and cirrus clouds from our dataset. We This package applies linear transformations to estimate surface reflec­
also discarded images in which clouds obscured an in situ sampling tance, incorporating approximations of aerosol optical thickness and
location. We then implemented an atmospheric correction algorithm total columnar water vapor using Bayesian statistics and the Copernicus
using the Satellite Invariant Atmospheric Correction (SIAC) package to Atmospheric Monitoring Service (Yin et al., 2019). We felt confident in

Fig. 3. A time series of the four indices

used in this study applied to all cloud-
free Sentinel-2 images during the study
period (October 2015 - December 2020).
NDVI = Normalized Difference Vegeta­
tion Index, NDCI = Normalized Differ­
ence Chlorophyll Index, B8AB4 = Band
8A Band 4 (narrow NIR and red bands),
and B3B2 = Band 3 Band 2 (green and
blue bands). All values are based on
atmospherically-corrected satellite data.
Black lines on the x-axis indicate sam­
pling dates at each in situ sampling
location while gray lines on the x-axis
indicate months with no in situ sam­
pling. (For interpretation of the refer­
ences to color in this figure legend, the
reader is referred to the web version of
this article.)

4
C. Kislik et al. Ecological Indicators 140 (2022) 109041

Fig. 4. Time series per spectral index (rows) and in situ sampling location (columns) from October 2015 to December 2020. This demonstrates the range of values
among the four indices tested in this study and how they vary per site.

using this atmospheric correction algorithm because Lobo et al. (2021) research that uses the red and NIR bands for greenness detection (Ma
successfully applied the SIAC algorithm to Sentinel-2 imagery within et al., 2021), (2) the Normalized Difference Chlorophyll Index (NDCI),
Google Earth Engine to derive NDCI values over inland waters. Their which uses Sentinel-2′ s first vegetation red-edge and red bands and
results demonstrated that this algorithm correlates highly (Pearson R = performs well in chl-a detection of hypereutrophic water bodies (Ca­
0.96) with the European Space Agency’s Sentinel-2 Surface Reflectance ballero et al., 2020; Mishra and Mishra, 2012; Watanabe et al., 2018),
product, which is available after 2019 (Lobo et al., 2021). After the (3) the Band 8A Band 4 (B8AB4) index, which was inspired by studies
initial image selection, image matching within 5 days of in situ collec­ developed to examine the differences between the red and red-edge
tion dates, cloud masking, and atmospheric correction, we were left with wavelengths in Sentinel-2 measurements (Khalili and Hasanlou,
a total of 33 Sentinel-2 images that were incorporated into analysis. 2019), and (4) the Band 3 Band 2 (B3B2) index, which uses the green
To assess which spectral index best detects chl-a at each sampling and blue bands for chl-a identification (Cillero Castro et al., 2020). Each
location, we applied the four spectral indices (Table 1) to the pre- of the wavelengths used within these indices has previously been tested
processed 33 Sentinel-2 images in Google Earth Engine. Each of these and proven to be effective in chl-a identification. We selected these
indices was calculated from the 12 bands available in Sentinel-2 imag­ specific bands for algal bloom detection because they have been tradi­
ery, including wavelengths ranging from the blue to the shortwave tionally used for chlorophyll-a identification; blue and green wavelength
infrared regions of the electromagnetic spectrum (Toming et al., 2016). indices operate well in clear bodies of water while red, red-edge, and
To identify chl-a, we used bands 2 (blue), 3 (green), 4 (red), 5 (vege­ near-infrared wavelength indices are successful in turbid waters because
tation red-edge), 8 (near-infrared), and 8A (narrow near-infrared) they are less affected by surrounding detritus and organic matter
(Wang et al., 2016). First, we extracted spectral index values from (Gilerson et al., 2010; Mishra and Mishra, 2012). Due to the variable
Sentinel-2 imagery by clipping these images to a 15 m circular buffer nature of these reservoirs, we explored both types of indices for this
that surrounded each in situ sampling location (Fig. 1). Buffers were study.
shifted by up to 15 m from the original in situ sampling location when The temporal resolution of Sentinel-2 imagery can greatly supple­
objects in the image, such as bridges or an intake barrier curtain, ment monthly in situ samples. As the Klamath reservoirs are dynamic
obscured water visibility and impacted index reflectance values. aquatic systems that have a legacy of nutrient inputs and retention
The four spectral indices tested were: (1) the Normalized Difference (Oliver et al., 2014), algal bloom proliferations in these bodies of water
Vegetation Index (NDVI), which is a ubiquitous index in terrestrial are very heterogeneous in space and time. To understand this

5
C. Kislik et al. Ecological Indicators 140 (2022) 109041

Fig. 5. Cubic polynomial regression (a) adjusted R2 and (b) Root Mean Square Error (RMSE) Values of the four Sentinel-2 spectral indices compared to in situ
chlorophyll-a data. Green signifies favorable values indicating high predictive potential (top 30th percentile adj. R2 values, bottom 30th percentile RMSE values in
µg/l). Statistical significance (*) for adj. R2 values is determined by: (*) p < 0.1; (**) p < 0.05; (***) p < 0.01; (****) p < 0.001. (For interpretation of the references to
color in this figure legend, the reader is referred to the web version of this article.)

complexity, we first collected all available Sentinel-2 images at each of Cyanobacteria Season (July 1 through October 31 for each year), as the
the three sampling locations and identified those coinciding with in situ peak bloom season in the Klamath River Basin occurs in this time period
sampling days. Next, we calculated the values of four spectral indices (Otten et al., 2015). Wet and dry years were categorized based on years
from the imagery on those dates: NDVI, NDCI, B8AB4, and B3B2 (Fig. 3). that experienced more than average (wet) or less than average (dry)
Fig. 3 shows this heterogeneity in that the Copco and Below Copco sites river discharge (mean = 1243 cubic feet per second) based on the 5-year
typically experience their highest chl-a values (as retrieved by Sentinel-2 average (2015–2020) discharge recorded at the United States Geological
spectral indices) in the spring and summer, while Iron Gate experiences Survey Klamath River Below John C. Boyle PowerPlant near Keno,
its highest chl-a measurements in the summer and fall. Winter months, Oregon, sampling station (USGS 11510700), which is upstream of all
when in situ sampling is halted, are typically a time of low chl-a values. sites. We then fit a cubic polynomial function (y = A + BX + CX2 + DX3
Four spectral indices applied to the Sentinel-2 imagery were explored and translated in this paper to y = index + chl-a + chl-a2 + chl-a3) to the
further in this paper (Fig. 3). paired chl-a value and spectral index values, resulting in 48 models total
(4 subgroups * 4 indices * 3 sites). We evaluated the significance of each
2.4. Data analysis relationship based on the p-value. We also calculated the root mean
square error (RMSE) and coefficient of determination, adjusted for the
We examined the relationship between in situ chl-a measurements number of terms in the model (adjusted R2), by comparing model pre­
and the spectral index values using cubic polynomial regression models dictions with observed chl-a values (Tehrani et al., 2021).
through the generalized linear model method within the stats package To examine higher and lower than average spectral index values per
and visualized using the ggplot2 package of the R program (R Core cyanobacteria season in the reservoirs, we selected the two highest-
Team, 2020; Wickham, 2016). We used this regression equation because performing indices, NDVI and NDCI, and calculated their annual stan­
of the non-linear relationship between the spectral reflectance and chl-a dardized anomalies in Google Earth Engine. Anomalies are commonly
values (Kwon et al., 2018; Martinez et al., 2020; Sharma et al., 2019) used in environmental science, such as for climate and drought moni­
and its success in related studies (Konik et al., 2020; O’Reilly et al., toring, to identify large departures from a long term mean (Anyamba
1998). We first categorized chl-a measurements into four temporal et al., 2001; Helama et al., 2009). This analysis is helpful in determining
subgroups per sampling site: Overall (all values per time series), Wet where blooms are occurring throughout the cyanobacteria season to
Years (2017 and 2019), Dry Years (2015, 2016, 2018, 2020), and better monitor temporal patterns of these events. To calculate the

Fig. 6. Regressions of the best-performing spectral index per sampling location based on the relationship between Sentinel-2 pixels and chlorophyll-a in situ data at
(a) Copco (Wet Year, NDCI), (b) Below Copco (Wet Year, NDCI), and (c) Iron Gate (Wet Year, NDCI). Gray shading: 95% confidence intervals; red lines demonstrate the
mean relationships for the polynomial (x3) regressions. R2 values represent adjusted R2 values. (For interpretation of the references to color in this figure legend, the
reader is referred to the web version of this article.)

6
C. Kislik et al. Ecological Indicators 140 (2022) 109041

Fig. 7. Annual standardized anomaly maps during the cyanobacteria season (July 1 - October 31) for the two most reliable indices tested in this study, NDVI and
NDCI. An annual Standardized Anomaly depicts higher than average and lower than average Sentinel-2 spectral index values per cyanobacteria season per year
(2016–2020) to identify areas that likely have more and less than average chl-a.

anomalies, we first organized the Sentinel-2 imagery into yearly com­ years (adj. R2 = 0.782, RMSE = 2 µg/l, p-Value < 0.01) (Fig. 5). In
posites (2016–2020) and then filtered these data to only include values comparison to in situ chl-a measurements, 19 of the 48 evaluated
from July 1 through October 31 each year, as this includes the annual regression results had moderate to strong relationships (adj. R2 > 0.4)
cyanobacteria season in this system (Jacoby and Kann, 2007; Otten and 26 of the 48 regression results were statistically significant (p-Value
et al., 2015). Next, we subtracted the overall mean per index from each < 0.1) (Fig. 5). Overall, all indices performed best at Copco (11 of 16
yearly composite value per index. Finally, we divided this resulting total regression results were statistically significant with p-Value < 0.1
value by the standard deviation for all years per index. This enabled us and 9 of 16 total results had moderate to strong relationships with adj.
to calculate the Z-score of the dataset, which is a metric that describes R2 > 0.4). NDCI was the most successful index across all sites and models
how many standard deviations an observation is from the mean (Patel (11 of 12 total regression results were statistically significant with p-
et al., 2007). For this study, we call the Z-score the standardized Value < 0.1 and 10 of 12 total results had moderate to strong re­
anomaly. The equation for this calculation is: [Z-Score = (Individual lationships with adj. R2 > 0.4), particularly at Copco and Iron Gate.
Year - Mean) / Standard Deviation of all Years]. The annual Standard­ B3B2 had the second-highest number of statistically significant models
ized Anomaly is used to decipher locations within each reservoir that (6 of 12 with p-Value < 0.1), performing best at Iron Gate, although
diverged from average Sentinel-2 spectral index values, determined per relationships were not as strong across all models (3 of 12 results with
year and during the cyanobacteria season (July - October). Higher than adj. R2 > 0.4). For this reason, NDVI appeared to perform the second-
average values can be used as a proxy to imply higher chl-a values, while best across all sites, after NDCI, with 4 of 12 moderate to strong re­
lower than average values serve as a proxy for decreased quantities of lationships (adj. R2 > 0.4) and 5 of 12 statistically-significant values (p-
chl-a. Value < 0.1). B8AB4 and NDVI performed similarly to one another
across several sites, with high statistical significance (p-Value < 0.01)
3. Results and relatively strong relationships at Copco in the categories of Overall
(entire time series), Copco Wet Year (2017 & 2019), and Copco Cya­
3.1. Satellite-based modeling of Chl-a nobacteria Season (July - October) (adj. R2 > 0.45 for NDVI and adj. R2
> 0.25 for B8AB4). However, error was higher at these sites for NDVI
Results revealed the seasonal and interannual heterogeneity of algal and B8AB4 than in other locations (as high as 20%). Of all the indices,
blooms in Copco and Iron Gate Reservoirs as detected by spectral indices B8AB4 performed the worst across all sites. NDCI and B3B2 had the
applied to Sentinel-2 satellite imagery. In general, algae appeared to be lowest RMSE values across all sites (below 9%) (Fig. 5). In terms of the
highest in the spring and summer months, with the highest peaks four temporal subgroups, the Overall category (entire time series) per­
observed in 2019 at Copco among all spectral indices, in 2019 for NDCI formed best (9 of 12 with p-Value < 0.1 and 5 of 12 with adj. R2 > 0.4),
and B3B2 values and 2020 for NDVI and B8AB4 values at Below Copco, followed by the Wet Years (2017 & 2019) (6 of 12 with p-Value < 0.1
and in 2018 for NDCI and B3B2 values and 2020 for NDVI and B8AB4 and 6 of 12 with adj. R2 > 0.4), the Cyanobacteria Season (July - Oct) (6
values at Iron Gate (Fig. 4). At each of the three sites, 2016 and 2017 of 12 with p-Value < 0.1 and 5 of 12 with adj. R2 > 0.4), and finally the
appeared to have generally lower spectral index values (Fig. 4). NDVI, Dry Years (2015, 2016, 2018, 2020) (5 of 12 with p-Value < 0.1 and 3 of
NDCI, and B8AB4 registered their lowest values at Iron Gate, and both 12 with adj. R2 > 0.4). Results indicate that pixel values that are
NDVI and B8AB4 experienced their highest values at Below Copco on collected within a 15-meter buffer around in situ sampling locations
average. NDCI had its highest values at Copco, with generally highest generally correspond well to in situ chl-a measurements within 5 days of
values on average across all spectral indices. B3B2 was fairly consistent collection.
across the three sites, with highest values at Iron Gate (Fig. 4). Strong relationships were observed between satellite-derived spec­
NDCI was the most successful spectral index in estimating chl-a from tral indices and in situ chl-a data at each site (Fig. 6). Below Copco had
Sentinel-2 imagery at Below Copco during the wet years (adj. R2 = the highest relationship based on the cubic polynomial regression (NDCI
0.841, RMSE = 4 µg/l, p-Value < 0.001) and Iron Gate during the wet adj. R2 = 0.841), while Iron Gate (NDCI adj. R2 = 0.782) and Copco

7
C. Kislik et al. Ecological Indicators 140 (2022) 109041

(NDCI adj. R2 = 0.777) had very similar values. There is a non-linear chl-a rates in these reservoirs during wet years were likely due to a
relationship between NDCI and chl-a, especially for chl-a values above combination of increased nutrient availability from runoff (Huang et al.,
0.25 µg/l at Copco and Below Copco, and 0.05 µg/l at Iron Gate. While 2014), as well as sloughing and flushing of certain aquatic grazers
most points lie within the gray 95% confidence intervals for each loca­ (Beaver et al., 2013). In this study, the efficacy of spectral indices for
tion, there are about five remaining points per plot that can be consid­ algal bloom detection was generally much better in wet years than dry
ered outliers (Fig. 6). years because of the higher and more consistent chl-a concentrations.
The red-edge wavelength outperformed NIR and narrow NIR wave­
3.2. Standardized anomalies of Chl-a estimates lengths in algal bloom detection throughout this study. Although NDVI
leverages the NIR band, it was not as useful as NDCI because this
Annual standardized anomalies demonstrate the relative differences wavelength has a tendency to overestimate chl-a in high chl-a concen­
from average spectral index values across space and time (Fig. 7). Spe­ trations, possibly resembling the oversaturation effect of high forest
cifically, the two best-performing indices across all sites, NDCI and canopy cover in terrestrial research (Jiang et al., 2006). High NDVI
NDVI, were used to show the biggest deviations from average annual RMSE values could be linked to the increase of prediction error associ­
values during the cyanobacteria season (July 1 - October 31) from 2016 ated with large ranges of chl-a concentrations (<5 µg/l or > 70 µg/l)
to 2020. Fig. 7 indicates that 2019 experienced the largest increases (Lobo et al., 2021). Furthermore, B8AB4, which incorporates the narrow
(positive changes from average spectral index values) across both res­ NIR and red bands in a similar equation to NDVI (Table 1), was found to
ervoirs, especially in Copco reservoir, while 2018 shows less than be the least effective index in this study. This poor performance is ex­
average values across both reservoirs, particularly in Copco reservoir pected as Sentinel-2′ s narrow NIR band (B8A) has demonstrated success
and the upstream half of Iron Gate. However, also in 2018, Iron Gate particularly in terrestrial studies such as leaf area index estimates
experienced higher than average values in the downstream portion of (Kaplan and Rozenstein, 2021) and burn severity identification
the reservoir. Iron Gate also exhibited higher standardized anomaly (Fernández-Manso et al., 2016), while the red (B4) and red edge (B5)
values in 2017 and 2020. Index values in 2016 and 2017 appear to be bands have also been applied to aquatic studies (Ambrose-Igho et al.,
fairly consistent without major increases or decreases across both res­ 2021; Shi et al., 2022). Additionally, without a large enough difference
ervoirs, and NDVI and NDCI performed similarly to one another in reflectance and absorption values between the narrow NIR and red
throughout the time series. Finally, these maps demonstrate that wavelengths, spectral index values for B8AB4 become muted and
depending on the year, different locations within the reservoirs have generally underestimate chl-a even in high concentrations (Fig. 4). Re­
various trends in anomalies. For example, during the wet years (2017 sults from this study corroborate findings from similar studies (Caballero
and 2019), Copco experienced higher than average trends near the et al., 2020; Cillero Castro et al., 2020), which attribute the detection of
Copco in situ sampling location and closer to the edges of the reservoir, phytoplankton blooms to high reflectance peaks in the red-edge and
while Iron Gate had higher than average values closer to the Below near-infrared bands and low reflectance values in the red band (Cabal­
Copco in situ sampling location. During the dry years (2016, 2018, lero et al., 2020). This reflectance pattern is characteristic of floating
2020), there were generally lower than average values across both res­ cyanobacteria blooms that appear green to the eye, which resembles
ervoirs, but Iron Gate may have experienced higher than average values reflectance peaks in healthy terrestrial vegetation (Govender et al.,
closer to the Iron Gate in situ sampling location. Standardized anomaly 2007). Although Sentinel-2 sensors were originally created for land-
maps are useful in determining the seasonal, interannual, and localized based studies, findings from this research and other related studies
spatial patterns of satellite-derived algal bloom occurrences. (Caballero et al., 2020; Cao et al., 2021; Cillero Castro et al., 2020; Lobo
et al., 2021) further support the notion that this satellite is appropriate
4. Discussion and can be adapted to aquatic studies.
Indices involving the blue and green portions of the electromagnetic
4.1. Sentinel-2 capabilities for algal bloom monitoring spectrum are often best for clearer waters with less organic material
(Alawadi, 2010; Binding et al., 2018). In this study, B3B2 performed
This research demonstrated that Sentinel-2 imagery is effective in poorly in terms of its relationship to in situ data, but had fairly low RMSE
detecting algal blooms in small freshwater reservoirs. NDCI best pre­ values (Fig. 5). This index proved to be moderately successful (adj. R2 >
dicted chl-a concentrations in the two reservoirs in relation to in situ 0.4) at Iron Gate sampling location (Fig. 5), most likely due to clarity of
sampling data, as demonstrated by moderate to strong adj. R2 values, the blooms on the surface of the water in this location, but was unreli­
high statistical significance, and low RMSE values (Fig. 5). This index able at Copco and Below Copco sites because blue and green wave­
uses the red-edge wavelength, which was originally applied in marine lengths tend to be affected by confounding factors of chl-a, including
and coastal environments (Caballero et al., 2020; Mishra and Mishra, detritus and color dissolved organic matter (Mishra and Mishra, 2012),
2012) and has since been tested and proven to effectively identify chl-a which are common in these extremely productive reservoirs. B3B2 both
in a variety of freshwater environments (Beck et al., 2017; Caballero overestimated chl-a in low concentrations (as demonstrated in similar
et al., 2020; Cillero Castro et al., 2020; Lobo et al., 2021; Weber et al., studies (Cillero Castro et al., 2020)) and underestimated chl-a in high
2020). Red-edge and NIR wavelengths are less affected by detritus and concentrations (Fig. 4). This can be linked to the pigment packaging
suspended particles (Xu et al., 2019), which are typical components of effect (Alcântara et al., 2016), which is a flattening effect that describes
Copco and Iron Gate reservoirs (Oliver et al., 2014; Otten et al., 2015), lower chl-a reflectance values under scenarios of high in situ chl-a
and helps explain why NDCI (and NDVI) performed well at a majority of concentrations (Stuart et al., 1998). B3B2 appeared to perform best
the sampling locations. NDCI was also able to capture peak chl-a con­ during the dry years (2016, 2018, and 2020) when cyanobacteria levels
centrations observed at each site (Fig. 4), consistent with previous are often reduced because of decreased nutrient influx. Although indices
studies that have shown the index to accurately detect elevated chl-a that use the blue and green can penetrate the water’s surface slightly
concentrations (Gilerson et al., 2010; Mishra and Mishra, 2012). The farther than indices in the NIR, red-edge, and red wavelengths (Legleiter
highest detection of chl-a observed in this study (3270 µg/l) greatly et al., 2011), it is likely that blue-green indices are more accurate when
influenced the regression model fit for Copco and likely led to low RMSE blooms are clearly visible directly on the water’s surface instead of
values (Fig. 5). It is important to note that there are only a few very high layered in higher concentrations below the surface due to an inability to
in situ chl-a measurements per sampling location that anchor the upper capture the full concentration. Therefore, B3B2 is most useful when
ends of the regression equations. NDCI was especially successful in applied to locations that typically have lower chl-a concentrations (<5
identifying blooms during the wet years (2017 and 2019) (Fig. 5), which µg/l), when blooms look bright green or blue to the human eye, and
included the most elevated chl-a values throughout the time series. High when blooms proliferate in the top layer of the water column.

8
C. Kislik et al. Ecological Indicators 140 (2022) 109041

4.2. Limitations of Sentinel-2 Chl-a detection satellite imagery could also be helpful for public health warnings. For
example, at our study sites, chl-a measurements exceeded the “Danger”
There are several limitations associated with this research. The pri­ category for Trigger Levels for Human & Animal Health in California
mary concern relates to the temporal resolution mismatch inherent to (United States Environmental Protection Agency, 2019) on at least 10
the in situ and satellite data comparison. Since in situ data were only separate days (Fig. 3). Increased monitoring of the Klamath reservoirs
collected once a month at each of the sampling locations, it was difficult through the use of open-access satellite imagery can help communities
to find Sentinel-2 satellite imagery that was both unobstructed by clouds prepare for and respond to these toxic events.
and within several days of the sampling day (Fig. 3). Therefore, the five- This study provides information on algal biomass estimations prior
day time difference between the in situ measurements and some of the to one of the largest dam removals in history (Allen, 2010). As Klamath
satellite data may cause uncertainty in the analysis, and similar studies River dam removal is expected to improve overall water quality and
attributed error to three-day discrepancies between in situ and Sentinel- diminish long term toxic bloom impacts in this system (Otten et al.,
2 data (Cillero Castro et al., 2020). However, regressions run separately 2015), results from this study provide baseline data that can help
on satellite data captured fewer than three days from in situ measure­ managers prepare for ecological restoration projects in future dam
ments did not render better results in our preliminary analyses. As algal removal scenarios. Also, as the dams are drawn down for removal,
blooms can appear and vanish within hours (Lee et al., 2005), it is spectral indices, especially NDCI and NDVI, applied to high-resolution
preferable to gather data on an hourly or daily timescale. Fig. 3 dem­ satellite imagery will be effective as a tool for on-going water quality
onstrates that once a strong relationship is established between satellite monitoring.
imagery and in situ data, satellite imagery can be used to fill temporal Finally, the benefits of augmenting sampling with satellite imagery
gaps and understand the large heterogeneity that is present in this sys­ analysis rather than only increasing field sampling are clear since there
tem. Missions such as PlanetScope (3 m spatial resolution) and RapidEye is limited money and time for additional in situ sampling efforts. Instead,
(5 m spatial resolution) offer daily imagery (Hu, 2021) and indices such strong relationships between spectral indices and in situ data can be
as NDVI and NDCI could potentially be calculated from these images to established, allowing managers to leverage satellite-derived products
detect algal blooms at even finer resolutions. Algal bloom satellite (Fig. 7) to expand the comprehension of the spatial and temporal dy­
detection could also be improved by harmonizing imagery from namics of algal blooms in these systems. This study demonstrated that
Sentinel-2 and Sentinel-3 (and potentially with high-resolution Planet NDCI is a useful tool to understand cyanobacteria dynamics and patterns
imagery (3–5 m)) to leverage the distinct spatial, temporal, and spectral in small reservoirs and lakes, which can be implemented to inform
resolution from each mission (Caballero et al., 2020; Pahlevan et al., localized management and mitigation strategies. Satellite imagery al­
2020). This could help reduce the mismatch between in situ and satellite lows an integration of understanding of algal bloom dynamics over
data and enable more precise monitoring of bloom dynamics in small space and time and thus provides a greater perspective on the variability
reservoirs. of the aquatic system than methods relying only on monthly in situ
Another limitation involves the detection of chl-a as a determinant observations.
for cyanobacteria species and toxicity. While some satellite missions
have spectral bands that can identify the presence of phycocyanin, a key 5. Conclusions
accessory pigment in cyanobacteria that can potentially produce toxins
such as microcystin (Beck et al., 2017), Sentinel-2 does not have this We determined that Sentinel-2 imagery can be used to improve
specificity in band wavelength. In contrast, the Sentinel-3 Ocean and detection of blooms at high spatial and temporal resolution and inform
Land Color Instrument sensor has a band centered at 620 nm that has algal bloom management and determination of baseline water quality
been used for more precise identification of cyanobacteria blooms conditions before dam removal. By quantifying the relationship between
(Ogashawara, 2019). However, these images are captured at coarser Sentinel-2 imagery and in situ chl-a data through the use of red-edge and
spatial resolution (300 m) and are thus less useful for small water bodies. NIR spectral indices, it is possible to track the seasonal, interannual, and
Although chl-a is useful in ascertaining algal biomass, the concentration local patterns of blooms in small reservoirs. NDCI proved to be the
of chl-a can only act as a proxy for blooms and indicates neither the optimal spectral index in this environment. We recommend that future
species nor the toxicity level of the bloom (Hallegraeff et al., 2004; studies use high-resolution satellite imagery that coincide in space and
Stumpf et al., 2003). However, toxin levels in this study have a high time with in situ water quality sampling during peak bloom events. This
correlation with chl-a concentrations (Fig. A.2), thus supporting these research promoted a greater understanding and communication of a
methods to detect both chl-a and microcystin toxin concentrations using highly dynamic system through the incorporation of a continuous sat­
Sentinel-2 imagery in these reservoirs. To remotely sense algal species or ellite mission and high-performance cloud computing via Google Earth
genera information, hyperspectral imagery at high spatial resolutions Engine, where code and analysis can be shared among researchers and
with daily return intervals would be optimal, although no current sat­ water quality managers. Research in this field can benefit from increased
ellite missions fit this objective. spatial, temporal, and spectral resolution, and upcoming hyperspectral
missions such as NASA’s Plankton, Aerosol, Cloud, ocean Ecosystem
4.3. Algal bloom management implications (PACE) paired with in situ data, and high spatial resolution imagery
from Sentinel and Planet missions can increase the ability to document
Despite the above mentioned limitations, Sentinel-2 chl-a time series and address future blooms. Finally, as dam removal is anticipated to
analysis provides a promising tool for expanding the spatial and tem­ reduce toxic blooms in the Klamath River Basin over time (Otten et al.,
poral measurements of algal blooms in small reservoirs. Currently in 2015), results from this study allow researchers to estimate algal bloom
Iron Gate Reservoir, PacifiCorp uses a physical barrier or “curtain” to risk aversion in future dam removal projects. Similar methods can be
curtail the downstream drift of toxic algae to the rest of the Klamath applied to a variety of freshwater environments, reducing the impact of
River (Stillwater Sciences, 2020). Enhanced monitoring of blooms with potentially toxic algal blooms in small lakes, rivers, and reservoirs
the use of Sentinel-2 imagery could help inform when barriers should be around the world.
installed or removed. Furthermore, the use of Google Earth Engine time
series information and standardized anomaly images (Fig. 7) can help to Funding
identify specific locations that may require additional sampling or
bloom mitigation measures, such as phosphate reduction (Paerl et al., This material is based upon work supported by the National Science
2018) or cyanobacteria removal and inactivation techniques (Westrick Foundation Graduate Research Fellowship Program under Grant No.
et al., 2010). More frequent assessment of algal blooms enabled by 1752814.

9
C. Kislik et al. Ecological Indicators 140 (2022) 109041

CRediT authorship contribution statement Data Availability:

Chippie Kislik: Conceptualization, Data curation, Methodology, Code based in Google Earth Engine (two scripts to analyze and
Software, Formal analysis, Validation, Project administration, Writing – display Sentinel-2 data) and R (two scripts to statistically analyze
original draft, Writing – review & editing. Iryna Dronova: Writing – Sentinel-2 and in situ data) can be found here: https://github.
review & editing. Theodore E. Grantham: Writing – review & editing. com/chippiekizzle/Klamath_Reservoirs.
Maggi Kelly: Conceptualization, Validation, Project administration,
Writing – original draft, Writing – review & editing. Acknowledgement

Declaration of Competing Interest The authors would like to thank the Karuk and Yurok tribes for their
support, and Dr. Jake Kann and Eli Asarian for invaluable feedback on
The authors declare that they have no known competing financial interpretation of results.
interests or personal relationships that could have appeared to influence
the work reported in this paper.
Appendix

Table A1
A list of acronyms mentioned in this manuscript.
B3B2 Band 3 Band 2
B8AB4 Band 8A Band 4
Chl-a Chlorophyll-a
NDCI Normalized Difference Chlorophyll Index
NDVI Normalized Difference Vegetation Index
NIR Near-infrared
RMSE Root Mean Squared Error

Fig. A1. In situ chlorophyll-a and microcystin values (µg/l) per sampling location from 2015 to 2020, measured by PacifiCorp.

Fig. A2. Linear regressions of in situ microcystin toxin and chlorophyll-a data at (a) Copco, (b) Below Copco, and (c) Iron Gate. Gray shading: 95% confidence
intervals; red lines demonstrate the mean relationships for the regression. R2 values represent adjusted R2 values.

10
C. Kislik et al. Ecological Indicators 140 (2022) 109041

References Dogliotti, A.I., Gossn, J.I., Vanhellemont, Q., Ruddick, K.G., 2018. Detecting and
Quantifying a Massive Invasion of Floating Aquatic Plants in the Río de la Plata
Turbid Waters Using High Spatial Resolution Ocean Color Imagery. Remote Sensing
Alawadi, F., 2010. Detection of surface algal blooms using the newly developed
10, 1140.
algorithm surface algal bloom index (SABI). Remote Sensing of the Ocean, Sea Ice,
Fernández-Manso, A., Fernández-Manso, O., Quintano, C., 2016. SENTINEL-2A red-edge
and Large.
spectral indices suitability for discriminating burn severity. Int. J. Appl. Earth Obs.
Alcântara, E., Watanabe, F., Rodrigues, T., Bernardo, N., 2016. An investigation into the
Geoinf. 50, 170–175.
phytoplankton package effect on the chlorophyll-a specific absorption coefficient in
Gilerson, A.A., Gitelson, A.A., Zhou, J., Gurlin, D., Moses, W., Ioannou, I., Ahmed, S.A.,
Barra Bonita reservoir. Brazil. Remote Sens. Lett. 7, 761–770.
2010. Algorithms for remote estimation of chlorophyll-a in coastal and inland waters
Allen, D.N., 2010. The Klamath hydroelectric settlement agreement: federal law, local
using red and near infrared bands. Opt. Express 18, 24109–24125.
compromise, and the largest dam removal project in history. Hastings W. -Nw. J.
Govender, M., Chetty, K., Bulcock, H., 2007. A review of hyperspectral remote sensing
Envt’l L. & Pol’y 16, 427.
and its application in vegetation and water resource studies. 1 33. https://doi.org/
Ambrose-Igho, G., Seyoum, W.M., Perry, W.L., O’Reilly, C.M., 2021. Spatiotemporal
10.4314/wsa.v33i2.49049.
Analysis of Water Quality Indicators in Small Lakes Using Sentinel-2 Satellite Data:
Hallegraeff, G.M., 1993. A review of harmful algal blooms and their apparent global
Lake Bloomington and Evergreen Lake, Central Illinois, USA. Environmental
increase. Phycologia 32, 79–99.
Processes 8, 637–660.
Hallegraeff, G.M., Anderson, D.M., Cembella, A.D., Enevoldsen, H.O., 2004. Manual on
Anderson, D.M., Cembella, A.D., Hallegraeff, G.M., 2012. Progress in understanding
harmful marine microalgae. Unesco.
harmful algal blooms: paradigm shifts and new technologies for research,
Ha, N.T.T., Thao, N.T.P., Koike, K., Nhuan, M.T., 2017. Selecting the Best Band Ratio to
monitoring, and management. Ann. Rev. Mar. Sci. 4, 143–176.
Estimate Chlorophyll-a Concentration in a Tropical Freshwater Lake Using Sentinel
Ansper, A., Alikas, K., 2018. Retrieval of Chlorophyll a from Sentinel-2 MSI Data for the
2A Images from a Case Study of Lake Ba Be (Northern Vietnam). ISPRS Int. J. Geo-
European Union Water Framework Directive Reporting Purposes. Remote Sensing
Inf. https://doi.org/10.3390/ijgi6090290.
11, 64.
Harke, M.J., Steffen, M.M., Gobler, C.J., Otten, T.G., Wilhelm, S.W., Wood, S.A., Paerl, H.
Anyamba, A., Tucker, C.J., Eastman, J.R., 2001. NDVI anomaly patterns over Africa
W., 2016. A review of the global ecology, genomics, and biogeography of the toxic
during the 1997/98 ENSO warm event. Int. J. Remote Sens. 22, 1847–1860.
cyanobacterium, Microcystis spp. Harmful Algae 54, 4–20.
Asarian, E., Kann, J., 2011. Phytoplankton and Nutrient Dynamics in Iron Gate and
Helama, S., Meriläinen, J., Tuomenvirta, H., 2009. Multicentennial megadrought in
Copco Reservoirs 2005- 2010. Prepared by Kier Associates and Aquatic Ecosystem
northern Europe coincided with a global El Niño-Southern Oscillation drought
Sciences for the Klamath Basin Tribal Water Quality Work Group.
pattern during the Medieval Climate Anomaly. Geology 37, 175–178.
Asarian, E., Kann, J. and W. Walker, 2009. Multi-year Nutrient Budget Dynamics for Iron
Ho, L.T., Goethals, P.L.M., 2019. Opportunities and Challenges for the Sustainability of
Gate and Copco Reservoirs, California. Prepared by Riverbend Sciences, Kier
Lakes and Reservoirs in Relation to the Sustainable Development Goals (SDGs).
Associates, Aquatic Ecosystem Sciences, and William Walker for the Karuk Tribe
Water. https://doi.org/10.3390/w11071462.
Department of Natural Resources, Orleans, CA.
Huang, T., Li, X., Rijnaarts, H., Grotenhuis, T., Ma, W., Sun, X., Xu, J., 2014. Effects of
Backer, L.C., Landsberg, J.H., Miller, M., Keel, K., Taylor, T.K., 2013. Canine Cyanotoxin
storm runoff on the thermal regime and water quality of a deep, stratified reservoir
Poisonings in the United States (1920s–2012): Review of Suspected and Confirmed
in a temperate monsoon zone, in Northwest China. Sci. Total Environ. 485–486,
Cases from Three Data Sources. Toxins 5, 1597–1628.
820–827.
Beaver, J.R., Jensen, D.E., Casamatta, D.A., Tausz, C.E., Scotese, K.C., Buccier, K.M.,
Hu, C., 2021. Remote detection of marine debris using satellite observations in the visible
Teacher, C.E., Rosati, T.C., Minerovic, A.D., Renicker, T.R., 2013. Response of
and near infrared spectral range: Challenges and potentials. Remote Sens. Environ.
phytoplankton and zooplankton communities in six reservoirs of the middle Missouri
259, 112414.
River (USA) to drought conditions and a major flood event. Hydrobiologia 705,
Hu, C., 2009. A novel ocean color index to detect floating algae in the global oceans.
173–189.
Remote Sens. Environ. 113, 2118–2129.
Beck, R., Xu, M., Zhan, S., Liu, H., Johansen, R., Tong, S., Yang, B., Shu, S., Wu, Q.,
Jacoby, J.M., Kann, J., 2007. The occurrence and response to toxic cyanobacteria in the
Wang, S., Berling, K., Murray, A., Emery, E., Reif, M., Harwood, J., Young, J.,
Pacific Northwest. North America. Lake Reserv. Manag. 23, 123–143.
Martin, M., Stillings, G., Stumpf, R., Su, H., Ye, Z., Huang, Y., 2017. Comparison of
Jiang, Z., Huete, A.R., Chen, J., Chen, Y., Li, J., Yan, G., Zhang, X., 2006. Analysis of
Satellite Reflectance Algorithms for Estimating Phycocyanin Values and
NDVI and scaled difference vegetation index retrievals of vegetation fraction.
Cyanobacterial Total Biovolume in a Temperate Reservoir Using Coincident
Remote Sens. Environ. 101, 366–378.
Hyperspectral Aircraft Imagery and Dense Coincident Surface Observations. Remote
Jia, T., Zhang, X., Dong, R., 2019. Long-Term Spatial and Temporal Monitoring of
Sensing. https://doi.org/10.3390/rs9060538.
Cyanobacteria Blooms Using MODIS on Google Earth Engine: A Case Study in Taihu
Binding, C.E., Greenberg, T.A., McCullough, G., Watson, S.B., Page, E., 2018. An analysis
Lake. Remote Sensing 11, 2269.
of satellite-derived chlorophyll and algal bloom indices on Lake Winnipeg. J. Great
Kann, J., Corum, S., 2006. Summary of 2005 Toxic Microcystis aeruginosa Trends in
Lakes Res. 44, 436–446.
Copco and Iron Gate Reservoirs on the Klamath River, CA. Prepared by Aquatic
Bozarth, C.S., Schwartz, A.D., Shepardson, J.W., Colwell, F.S., Dreher, T.W., 2010.
Ecosystem Sciences and the Karuk Tribe Department of Natural Resources for the
Population turnover in a Microcystis bloom results in predominantly nontoxigenic
Karuk Tribe.
variants late in the season. Appl. Environ. Microbiol. 76, 5207–5213.
Kaplan, G., Rozenstein, O., 2021. Spaceborne Estimation of Leaf Area Index in Cotton,
Bramich, J., Bolch, C.J.S., Fischer, A., 2021. Improved red-edge chlorophyll-a detection
Tomato, and Wheat Using Sentinel-2. Land 10, 505.
for Sentinel 2. Ecol. Indic. 120, 106876.
Karlson, B., Cusack, C., Bresnan, E., 2010. Microscopic and molecular methods for
Bresciani, M., Cazzaniga, I., Austoni, M., Sforzi, T., Buzzi, F., Morabito, G., Giardino, C.,
quantitative phytoplankton analysis 303.
2018. Mapping phytoplankton blooms in deep subalpine lakes from Sentinel-2A and
Kenneth Hudnell, H., 2008. Cyanobacterial Harmful Algal Blooms: State of the Science
Landsat-8. Hydrobiologia 824, 197–214.
and Research Needs. Springer Science & Business Media.
Brooks, B.W., Lazorchak, J.M., Howard, M.D.A., 2016. Are harmful algal blooms
Khalili, M.H., Hasanlou, M., 2019. Harmful Algal Blooms monitoring using Sentinel-2
becoming the greatest inland water quality threat to public health and aquatic
satellite images. ISPRS - Int. Arch. Photogramm. Remote Sens. Spat. Inf. Sci. XLII-4/
ecosystems? Environmentalist.
W18, 609–613.
Caballero, I., Fernández, R., Escalante, O.M., Mamán, L., Navarro, G., 2020. New
Kislik, C., Dronova, I., Kelly, M., 2018. UAVs in Support of Algal Bloom Research: A
capabilities of Sentinel-2A/B satellites combined with in situ data for monitoring
Review of Current Applications and Future Opportunities. Drones 2, 35.
small harmful algal blooms in complex coastal waters. Sci. Rep. 10, 8743.
Klamath River Renewal Corporation, n.d. Lower Klamath Project Exhibit A-1: Definite
Caballero, I., Stumpf, R.P., 2020. Towards Routine Mapping of Shallow Bathymetry in
Decommissioning Plan (No. FERC No. 14803).
Environments with Variable Turbidity: Contribution of Sentinel-2A/B Satellites
Konik, M., Kowalczuk, P., Zabłocka, M., Makarewicz, A., Meler, J., Zdun, A., Darecki, M.,
Mission. Remote Sensing 12, 451.
2020. Empirical Relationships between Remote-Sensing Reflectance and Selected
Cao, M., Qing, S., Jin, E., Hao, Y., Zhao, W., 2021. A spectral index for the detection of
Inherent Optical Properties in Nordic Sea Surface Waters for the MODIS and OLCI
algal blooms using Sentinel-2 Multispectral Instrument (MSI) imagery: a case study
Ocean Colour Sensors. Remote Sensing 12, 2774.
of Hulun Lake, China. Int. J. Remote Sens. 42, 4514–4535.
Kudela, R.M., Stumpf, R.P., Petrov, P., 2017. Acquisition and analysis of remote sensing
Chorus, I., Falconer, I.R., Salas, H.J., Bartram, J., 2000. Health risks caused by freshwater
imagery of harmful algal blooms.
cyanobacteria in recreational waters. J. Toxicol. Environ. Health B Crit. Rev. 3,
Kwon, Y.S., Baek, S.H., Lim, Y.K., Pyo, J., Ligaray, M., Park, Y., Cho, K.H., 2018.
323–347.
Monitoring Coastal Chlorophyll-a Concentrations in Coastal Areas Using Machine
Cillero Castro, C., Domínguez Gómez, J.A., Delgado Martín, J., Hinojo Sánchez, B.A.,
Learning Models. Water 10, 1020.
Cereijo Arango, J.L., Cheda Tuya, F.A., Díaz-Varela, R., 2020. An UAV and Satellite
Laughrey, Z.R., Christensen, V.G., Dusek, R.J., Senegal, S., Lankton, J.S., Ziegler, T.A.,
Multispectral Data Approach to Monitor Water Quality in Small Reservoirs. Remote
Jones, L.C., Jones, D.K., Williams, B.M., Gordon, S., Clyde, G.A., Emery, E.B.,
Sensing 12, 1514.
Loftin, K.A., 2021. A review of algal toxin exposures on reserved federal lands and
Clark, J.M., Schaeffer, B.A., Darling, J.A., Urquhart, E.A., Johnston, J.M., Ignatius, A.R.,
among trust species in the United States. Crit. Rev. Environ. Sci. Technol. 1–24.
Myer, M.H., Loftin, K.A., Werdell, P.J., Stumpf, R.P., 2017. Satellite monitoring of
Lee, J.H.W., Hodgkiss, I.J., Wong, K.T.M., Lam, I.H.Y., 2005. Real time observations of
cyanobacterial harmful algal bloom frequency in recreational waters and drinking
coastal algal blooms by an early warning system. Estuar. Coast. Shelf Sci. 65,
water sources. Ecol. Indic. 80, 84–95.
172–190.
Cloern, J.E., 1996. Phytoplankton bloom dynamics in coastal ecosystems: A review with
Legleiter, C.J., Kinzel, P.J., Overstreet, B.T., 2011. Evaluating the potential for remote
some general lessons from sustained investigation of San Francisco Bay, California.
bathymetric mapping of a turbid, sand-bed river: 1. Field spectroscopy and radiative
Reviews of Geophysics. https://doi.org/.
transfer modeling. Water Resources Research. https://doi.org/10.1029/
Davidson, K., Gowen, R.J., Harrison, P.J., Fleming, L.E., Hoagland, P., Moschonas, G.,
2011wr010591.
2014. Anthropogenic nutrients and harmful algae in coastal waters. J. Environ.
Manage. 146, 206–216.

11
C. Kislik et al. Ecological Indicators 140 (2022) 109041

Lobo, F. de L., Nagel, G.W., Maciel, D.A., Carvalho, L.A.S. de, Martins, V.S., Barbosa, C.C. Shen, L., Xu, H., Guo, X., 2012. Satellite remote sensing of harmful algal blooms (HABs)
F., Novo, E.M.L. de M., 2021. AlgaeMAp: Algae Bloom Monitoring Application for and a potential synthesized framework. Sensors 12, 7778–7803.
Inland Waters in Latin America. Remote Sensing 13, 2874. Shi, J., Shen, Q., Yao, Y., Li, J., Chen, F., Wang, R., Xu, W., Gao, Z., Wang, L., Zhou, Y.,
Lopez, C.B., Jewett, E.B., Dortch, Q., Walton, B.T., Hudnell, H.K., 2008. Scientific 2022. Estimation of Chlorophyll-a Concentrations in Small Water Bodies:
assessment of freshwater harmful algal blooms. Comparison of Fused Gaofen-6 and Sentinel-2 Sensors. Remote Sensing 14, 229.
Maeda, E.E., Lisboa, F., Kaikkonen, L., Kallio, K., Koponen, S., Brotas, V., Kuikka, S., Stillwater Sciences, 2020. Environmental Impact Report for the Lower Klamath Project
2019. Temporal patterns of phytoplankton phenology across high latitude lakes License Surrender Volume III (No. State Clearinghouse No. 2016122047). Prepared
unveiled by long-term time series of satellite data. Remote Sens. Environ. 221, for State Water Resources Control Board, Division of Water Rights, Sacramento, CA.
609–620. Stuart, V., Sathyendranath, S., Platt, T., Maass, H., Irwin, B.D., 1998. Pigments and
Ma, J., Jin, S., Li, J., He, Y., Shang, W., 2021. Spatio-Temporal Variations and Driving species composition of natural phytoplankton populations: effect on the absorption
Forces of Harmful Algal Blooms in Chaohu Lake: A Multi-Source Remote Sensing spectra. J. Plankton Res. 20, 187–217.
Approach. Remote Sensing 13, 427. Stumpf, R.P., Culver, M.E., Tester, P.A., Tomlinson, M., Kirkpatrick, G.J., Pederson, B.A.,
Martinez, E., Gorgues, T., Lengaigne, M., Fontana, C., Sauzède, R., Menkes, C., Uitz, J., Di Truby, E., Ransibrahmanakul, V., Soracco, M., 2003. Monitoring Karenia brevis
Lorenzo, E., Fablet, R., 2020. Reconstructing Global Chlorophyll-a Variations Using a blooms in the Gulf of Mexico using satellite ocean color imagery and other data.
Non-linear Statistical Approach. Front. Mar. Sci. 7, 464. Harmful Algae 2, 147–160.
Matthews, M.W., Bernard, S., Robertson, L., 2012. An algorithm for detecting trophic Stumpf, R.P., Tomlinson, M.C., 2007. Remote sensing of harmful algal blooms, in:
status (chlorophyll-a), cyanobacterial-dominance, surface scums and floating Remote Sensing of Coastal Aquatic Environments. Springer, pp. 277–296.
vegetation in inland and coastal waters. Remote Sens. Environ. 124, 637–652. Tehrani, N.A., Janalipour, M., Babaei, H., 2021. Estimating Water Surface Chlorophyll-a
Mishra, S., Mishra, D.R., 2012. Normalized difference chlorophyll index: A novel model Concentration by Big Remote Sensing Data in the Persian Gulf, Bushehr. Remote
for remote estimation of chlorophyll-a concentration in turbid productive waters. Sensing in Earth Systems Sciences 4, 87–95.
Remote Sens. Environ. 117, 394–406. Toming, K., Kutser, T., Laas, A., Sepp, M., Paavel, B., Nõges, T., 2016. First Experiences in
Mishra, S., Stumpf, R.P., Schaeffer, B., Werdell, P.J., Loftin, K.A., Meredith, A., 2021. Mapping Lake Water Quality Parameters with Sentinel-2 MSI Imagery. Remote
Evaluation of a satellite-based cyanobacteria bloom detection algorithm using field- Sensing 8, 640.
measured microcystin data. Sci. Total Environ. 774, 145462. United States Environmental Protection Agency, 2019. Recommended Human Health
Moisander, P.H., Ochiai, M., Lincoff, A., 2009. Nutrient limitation of Microcystis Recreational Ambient Water Quality Criteria or Swimming Advisories for
aeruginosa in northern California Klamath River reservoirs. Harmful Algae 8, Microcystins and Cylindrospermopsin (No. EPA 822-R-19-001). U.S. Environmental
889–897. Protection Agency, Office of Water Health and Ecological Criteria Division,
Ogashawara, I., 2019. The Use of Sentinel-3 Imagery to Monitor Cyanobacterial Blooms. Washington, DC.
Environments 6, 60. U.S. Department of the Interior, U.S. Department of Commerce, National Marine
Oliver, A.A., Dahlgren, R.A., Deas, M.L., 2014. The upside-down river: Reservoirs, algal Fisheries Service, 2013. Klamath Dam Removal Overview - Report for the Secretary
blooms, and tributaries affect temporal and spatial patterns in nitrogen and of the Interior: An Assessment of Science and Technical Information (No. 1.1).
phosphorus in the Klamath River. USA. J. Hydrol. 519, 164–176. Vaičiūtė, D., Bučas, M., Bresciani, M., Dabulevičienė, T., Gintauskas, J., Mėžinė, J.,
O’Reilly, J.E., Maritorena, S., Mitchell, B.G., Siegel, D.A., Carder, K.L., Garver, S.A., Tiškus, E., Umgiesser, G., Morkūnas, J., De Santi, F., Bartoli, M., 2021. Hot moments
Kahru, M., McClain, C., 1998. Ocean color chlorophyll algorithms for SeaWiFS. and hotspots of cyanobacteria hyperblooms in the Curonian Lagoon (SE Baltic Sea)
J. Geophys. Res. 103, 24937–24953. revealed via remote sensing-based retrospective analysis. Sci. Total Environ. 769,
Otten, T.G., Crosswell, J.R., Mackey, S., Dreher, T.W., 2015. Application of molecular 145053.
tools for microbial source tracking and public health risk assessment of a Microcystis Walker, W.W., Walker, J.D., Kann, J., 2012. Evaluation of Water and Nutrient Balances
bloom traversing 300 km of the Klamath River. Harmful Algae 46, 71–81. for the Upper Klamath Lake Basin in Water Years 1992–2010. Technical Report to
Paerl, H.W., Fulton 3rd, R.S., Moisander, P.H., Dyble, J., 2001. Harmful freshwater algal the Klamath Tribes Natural Resources Department, Chiloquin, OR.
blooms, with an emphasis on cyanobacteria. ScientificWorldJournal 1, 76–113. Wang, L., Xu, M., Liu, Y., Liu, H., Beck, R., Reif, M., Emery, E., Young, J., Wu, Q., 2020.
Paerl, H.W., Otten, T.G., Kudela, R., 2018. Mitigating the Expansion of Harmful Algal Mapping Freshwater Chlorophyll-a Concentrations at a Regional Scale Integrating
Blooms Across the Freshwater-to-Marine Continuum. Environ. Sci. Technol. 52, Multi-Sensor Satellite Observations with Google Earth Engine. Remote Sensing 12,
5519–5529. 3278.
Pahlevan, N., Sarkar, S., Franz, B.A., Balasubramanian, S.V., He, J., 2017. Sentinel-2 Wang, Q., Shi, W., Li, Z., Atkinson, P.M., 2016. Fusion of Sentinel-2 images. Remote Sens.
MultiSpectral Instrument (MSI) data processing for aquatic science applications: Environ. 187, 241–252.
Demonstrations and validations. Remote Sens. Environ. 201, 47–56. Watanabe, F., Alcântara, E., Rodrigues, T., Rotta, L., Bernardo, N., Imai, N., 2018.
Pahlevan, N., Smith, B., Schalles, J., Binding, C., Cao, Z., Ma, R., Alikas, K., Kangro, K., Remote sensing of the chlorophyll-a based on OLI/Landsat-8 and MSI/Sentinel-2A
Gurlin, D., Hà, N., Matsushita, B., Moses, W., Greb, S., Lehmann, M.K., Ondrusek, M., (Barra Bonita reservoir, Brazil). An. Acad. Bras. Cienc. 90, 1987–2000.
Oppelt, N., Stumpf, R., 2020. Seamless retrievals of chlorophyll-a from Sentinel-2 Watercourse Engineering, Inc., 2020. Klamath River Baseline Water Quality Sampling -
(MSI) and Sentinel-3 (OLCI) in inland and coastal waters: A machine-learning 2019 Annual Report. Prepared for the KHSA Water Quality Monitoring Group.
approach. Remote Sens. Environ. 240, 111604. Weber, S.J., Mishra, D.R., Wilde, S.B., Kramer, E., 2020. Risks for cyanobacterial harmful
Patel, N.R., Chopra, P., Dadhwal, V.K., 2007. Analyzing spatial patterns of algal blooms due to land management and climate interactions. Sci. Total Environ.
meteorological drought using standardized precipitation index. Meteorol. Appl. 14, 703, 134608.
329–336. Weirich, C.A., Miller, T.R., 2014. Freshwater harmful algal blooms: toxins and children’s
Ranjbar, M.H., Hamilton, D.P., Etemad-Shahidi, A., Helfer, F., 2021. Individual-based health. Curr. Probl. Pediatr. Adolesc. Health Care 44, 2–24.
modelling of cyanobacteria blooms: Physical and physiological processes. Sci. Total Westrick, J.A., Szlag, D.C., Southwell, B.J., Sinclair, J., 2010. A review of cyanobacteria
Environ. 792, 148418. and cyanotoxins removal/inactivation in drinking water treatment. Anal. Bioanal.
Rastogi, R.P., Sinha, R.P., Incharoensakdi, A., 2014. The cyanotoxin-microcystins: Chem. 397, 1705–1714.
current overview. Rev. Environ. Sci. Biotechnol. 13, 215–249. Wickham, H., 2016. ggplot2: Elegant Graphics for Data Analysis. Springer.
R Core Team, 2020. R: A language and environment for statistical computing. Xu, M., Liu, H., Beck, R., Lekki, J., Yang, B., Shu, S., Kang, E.L., Anderson, R.,
Saberioon, M., Brom, J., Nedbal, V., Souc̆ ek, P., Císar̆ , P., 2020. Chlorophyll-a and total Johansen, R., Emery, E., Reif, M., Benko, T., 2019. A spectral space partition guided
suspended solids retrieval and mapping using Sentinel-2A and machine learning for ensemble method for retrieving chlorophyll-a concentration in inland waters from
inland waters. Ecol. Indic. 113, 106236. Sentinel-2A satellite imagery. J. Great Lakes Res. 45, 454–465.
Sebastiá-Frasquet, M.-T., Aguilar-Maldonado, J.-A., Herrero-Durá, I., Santamaría-del- Yin, F., Lewis, P.E., Gomez-Dans, J., Wu, Q., 2019. A sensor-invariant atmospheric
Ángel, E., Morell-Monzó, S., Estornell, J., 2020. Advances in the Monitoring of Algal correction method: Application to Sentinel-2/MSI and Landsat 8/OLI.
Blooms by Remote Sensing: A Bibliometric Analysis. NATO Adv. Sci. Inst. Ser. E Zong, J.-M., Wang, X.-X., Zhong, Q.-Y., Xiao, X.-M., Ma, J., Zhao, B., 2019. Increasing
Appl. Sci. 10, 7877. Outbreak of Cyanobacterial Blooms in Large Lakes and Reservoirs under Pressures
Sharma, P., Ueranantasun, A., Tongkumchum, P., Eso, M., 2019. Modelling of from Climate Change and Anthropogenic Interferences in the Middle-Lower Yangtze
Chlorophyll-a Concentration Patterns from Satellite Data Using Cubic Spline River Basin. Remote Sensing 11, 1754.
Function in Pattani Bay, Thailand. Nat. Environ. Pollut. Technol. 18.

12